CoDAC: Autonomous Obstacle Avoidance Optimization for Unmanned Surface Vehicle Clusters via Multi-Modal Dynamic Perception and Collaborative Community Detection

Autonomous obstacle avoidance for unmanned surface vehicle (USV) clusters in dynamic marine environments faces challenges including heterogeneous sensor fusion, multi-objective optimization conflicts, and scalable swarm coordination. This paper proposes CoDAC, an autonomous obstacle avoidance optimization method that synergizes multimodal dynamic perception and incremental community detection. The method establishes an integrated "perception-planning-collaboration" framework. First, cross-modal feature fusion and spatio-temporal alignment techniques significantly enhance environmental perception accuracy and robustness, effectively addressing heterogeneous sensor data fusion challenges. Second, an incremental dynamic community detection mechanism achieves adaptive task group partitioning for clusters, substantially reducing communication loads and computational complexity while ensuring high-efficiency collaboration at scale. Meanwhile, the Improved Velocity Obstacle Model (IVO-DWA) integrates Triangle Obstacle Zone (TOZ) prediction with multi-objective optimization, enabling real-time trade-offs among path length, smoothness, and compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). Simulation experiments demonstrate that CoDAC exhibits superior real-time performance and stability in complex scenarios, strictly adheres to maritime rules, and improves the emergency obstacle avoidance success rate by 26.7% (to 96.7%). The proposed method provides a highly reliable and scalable solution for the collaboration of unmanned systems in complex and dynamic marine environments.